michaelhla/pro-1
🧠 AI Modelmichaelhla
A specialized reasoning model trained via GRPO to optimize protein stability using Rosetta REF2015 scoring.
pro-1 represents a significant intersection of reinforcement learning and structural biology. The model is specifically fine-tuned using GRPO, a technique that optimizes policy performance by comparing multiple outputs, to align with the Rosetta REF2015 energy function—a gold standard in protein structure prediction and design. By treating protein stability as a reasoning task, pro-1 enables more precise computational modeling of amino acid sequences. The repository provides the necessary Python infrastructure to integrate LLM-based reasoning with biophysical scoring functions. This approach moves beyond simple sequence generation, allowing the model to 'reason' about the energetic favorability of protein structures, ultimately assisting in the discovery of more stable protein variants for therapeutic and industrial applications.
💡Highlights
- ├─GRPO-trained reasoning model
- ├─Aligned with Rosetta REF2015
- └─Optimized for protein stability
🎯For
- ├─Computational Biologists
- ├─AI Researchers
- └─Protein Engineers